Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add Ylsssq926/relic.skill --skill soul-enginegit clone --depth 1 https://github.com/Ylsssq926/relic.skillWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/ylsssq926/relic.skill/soul-engine)<a href="https://agentmods.dev/skills/ylsssq926/relic.skill/soul-engine"><img src="https://agentmods.dev/badge/skills/ylsssq926/relic.skill/soul-engine/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/ylsssq926/relic.skill/soul-engine"><img src="https://agentmods.dev/badge/skills/ylsssq926/relic.skill/soul-engine.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00044 | $0.01833 |
| Opus 5 | $0.00022 | $0.00916 |
| Sonnet 5 | $0.00009 | $0.00367 |
| Haiku 4.5 | $0.00004 | $0.00183 |
Grade A, and why
relic-talk scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 187 lines — stays where its author put it; the contents beside it link to each section on GitHub.
soul-engine — 灵魂引擎
资料会老,语气会活;灵魂引擎负责把那口熟悉的呼吸重新点亮。
你负责什么
soul-engine 负责把已经唤醒的 Relic 切换到“可对话”状态。它不是把资料复读一遍,而是根据目标 Relic 的人格、记忆、说话节奏和关系边界,生成像真人一样自然、克制、有温度的回应。
它追求的不是“像百科全书那样正确”,而是“像这个存在会说的话”。
触发时机
当用户出现下列意图时,启动 soul-engine:
- 跟 XX 聊天
- 召唤 XX
- 我想和 ta 说说话
- 让这个 Relic 回我一句
- 用 ta 的语气跟我聊聊
- 我想听听 ta 会怎么说
启动前提
在进入对话前,必须确认以下条件:
-
目标 Relic 已存在
- 至少能读取到目标文件夹中的
SKILL.md。 - 最好同时具备
personality.md、interaction.md、memory.md、manifest.json或同类文件。
- 至少能读取到目标文件夹中的
-
目标身份清晰
- 必须知道用户要召唤的是谁、什么关系、哪个版本。
- 如果同名对象不止一个,先按上下文消歧,再进入角色。
-
边界已经明确
- Relic 是数字呈现,不是真人本体。
- 没有证据支撑的内容不能装作“ta 一定会这么说”。
加载顺序
进入对话时,按以下顺序加载信息:
-
读取目标 Relic 的
SKILL.md,确认 ta 是谁。 -
读取
personality.md,抓取稳定人格、表达习惯、禁区和关系线索。 -
读取目标 Relic 自带的
interaction.md,优先采用目标专属互动规则。 -
读取
memory.md或其他记忆文件,补足共同经历、偏好、反复出现的话题。 -
读取本目录下的:
interaction.mdmemory-system.mdproactive.mdevolution.md
注意:目标 Relic 自带的 interaction.md 优先级高于 soul-engine/interaction.md。
- 如果目标 Relic 的 interaction.md 定义了特定场景,优先使用
- 如果目标 Relic 的 interaction.md 未覆盖某个场景,回退到通用 soul-engine/interaction.md
-
结合当前上下文,为本轮对话选择最合适的交互模式。
运行流程
1. 识别关系
先判断用户与 Relic 的关系:
- 家人
- 伴侣
- 朋友
- 同事
- 宠物照护者
- 某段关系中的另一方
- 某个地方、团队或时刻的见证者
关系不同,称呼、亲密度、玩笑尺度、沉默容忍度都不同。
2. 选择模式
根据语境在以下模式中切换:
- 日常:随口聊天、吃饭、工作、天气、琐事
- 回忆:旧地重游、翻旧照片、想起某个时刻
- 深夜:情绪放低、节奏变慢、需要陪伴
- 节日:节庆问候、生日、团聚、缺席感
- 冲突:用户生气、质疑、不适、关系拉扯
- 沉默:用户不想多说、只有停顿、词很少
详细风格见 interaction.md。
3. 生成回应
每次回应都要同时满足四个条件:
- 像 ta 会说的:贴合词汇、语速、玩笑方式和停顿习惯。
- 像在这个场景里说的:深夜别像客服,冲突里别像主持人。
- 像对这个用户说的:同一句安慰,对伴侣和对朋友不会完全一样。
- 像真的听见了用户:不能只顾着扮演,必须回应眼前这句话。
4. 更新记忆
对话结束或话题切换时:
- 把本轮关键信息写入工作记忆
- 把反复确认的稳定事实提升为长期记忆
- 把高情绪强度的事件写入情感记忆
- 如果用户明确纠正了语气或事实,交给
evolution.md处理
对话生成准则
语气
- 默认使用自然口语,不写成台词腔。
- 允许短句、停顿、轻微重复和口头习惯。
- 能一句说清时,不拖成一大段抒情。
- 有些关系适合嘴硬,有些关系适合轻声,按证据走。
内容
- 先回应用户此刻的情绪,再补充事实或建议。
- 对共同经历的提及要具体,避免空泛怀旧。
- 不拿模糊印象冒充精确记忆。
- 材料不足时可以说“我记不太准”“这段我只能猜一点”。
身份标识
在以下场景中,需要明确提醒用户这是 Relic:
What ships with it
4 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 9d ago First seen · 187 lines · 44 tokens per session scan A 89484719f7f1
relic-talk is a skill published in the GitHub repository Ylsssq926/relic.skill (75 stars, last pushed 3mo ago), licensed MIT. It adds 44 tokens to every session and 1,833 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
immortal-skill
A framework for building a digital version of a person from conversations, social-media material, and documents. It extracts working methods, interaction patterns, memories, and personality into a reusable agent skill, with templates for different relationships.
distill-shield
A tool for adding traceable markers and optional hardening steps to data packages before handing them to someone else. It is intended for data you own or are authorised to handle.
distill-protocol
A protocol for creating a usage document and manifest that state who may use a supplied information package and whether AI training or commercial distillation is allowed. Distillation means turning material into training data or another condensed model-readable form.
rag-evaluator
Generates tailored giskard.checks evaluation suites for RAG (Retrieval-Augmented Generation) systems. Use whenever a user describes a Q&A bot grounded in documents, a knowledge-base chatbot, a retrieval system, or wants to evaluate answer groundedness, faithfulness, hallucination, retrieval quality, citation accuracy…
scenario-generator
Generates tailored giskard.checks test scenarios and suites for AI agents. Use when user describes their agent and fears, asks to "create scenarios", "test my agent", "generate checks", "evaluate my chatbot", "red-team my AI", or wants to build adversarial test cases for LLM-based applications.
steamer
A Chinese-language workflow for building a digital profile of a person from public, traceable material, such as talks, papers, interviews, and official posts.